2 * ring buffer based function tracer
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 * Originally taken from the RT patch by:
8 * Arnaldo Carvalho de Melo <acme@redhat.com>
10 * Based on code from the latency_tracer, that is:
11 * Copyright (C) 2004-2006 Ingo Molnar
12 * Copyright (C) 2004 William Lee Irwin III
14 #include <linux/ring_buffer.h>
15 #include <linux/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/smp_lock.h>
21 #include <linux/notifier.h>
22 #include <linux/irqflags.h>
23 #include <linux/debugfs.h>
24 #include <linux/pagemap.h>
25 #include <linux/hardirq.h>
26 #include <linux/linkage.h>
27 #include <linux/uaccess.h>
28 #include <linux/kprobes.h>
29 #include <linux/ftrace.h>
30 #include <linux/module.h>
31 #include <linux/percpu.h>
32 #include <linux/splice.h>
33 #include <linux/kdebug.h>
34 #include <linux/string.h>
35 #include <linux/ctype.h>
36 #include <linux/init.h>
37 #include <linux/poll.h>
38 #include <linux/gfp.h>
42 #include "trace_output.h"
44 #define TRACE_BUFFER_FLAGS (RB_FL_OVERWRITE)
46 unsigned long __read_mostly tracing_max_latency;
47 unsigned long __read_mostly tracing_thresh;
50 * On boot up, the ring buffer is set to the minimum size, so that
51 * we do not waste memory on systems that are not using tracing.
53 int ring_buffer_expanded;
56 * We need to change this state when a selftest is running.
57 * A selftest will lurk into the ring-buffer to count the
58 * entries inserted during the selftest although some concurrent
59 * insertions into the ring-buffer such as trace_printk could occurred
60 * at the same time, giving false positive or negative results.
62 static bool __read_mostly tracing_selftest_running;
65 * If a tracer is running, we do not want to run SELFTEST.
67 bool __read_mostly tracing_selftest_disabled;
69 /* For tracers that don't implement custom flags */
70 static struct tracer_opt dummy_tracer_opt[] = {
74 static struct tracer_flags dummy_tracer_flags = {
76 .opts = dummy_tracer_opt
79 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
85 * Kill all tracing for good (never come back).
86 * It is initialized to 1 but will turn to zero if the initialization
87 * of the tracer is successful. But that is the only place that sets
90 static int tracing_disabled = 1;
92 DEFINE_PER_CPU(local_t, ftrace_cpu_disabled);
94 static inline void ftrace_disable_cpu(void)
97 local_inc(&__get_cpu_var(ftrace_cpu_disabled));
100 static inline void ftrace_enable_cpu(void)
102 local_dec(&__get_cpu_var(ftrace_cpu_disabled));
106 static cpumask_var_t __read_mostly tracing_buffer_mask;
108 /* Define which cpu buffers are currently read in trace_pipe */
109 static cpumask_var_t tracing_reader_cpumask;
111 #define for_each_tracing_cpu(cpu) \
112 for_each_cpu(cpu, tracing_buffer_mask)
115 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
117 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
118 * is set, then ftrace_dump is called. This will output the contents
119 * of the ftrace buffers to the console. This is very useful for
120 * capturing traces that lead to crashes and outputing it to a
123 * It is default off, but you can enable it with either specifying
124 * "ftrace_dump_on_oops" in the kernel command line, or setting
125 * /proc/sys/kernel/ftrace_dump_on_oops to true.
127 int ftrace_dump_on_oops;
129 static int tracing_set_tracer(const char *buf);
131 #define BOOTUP_TRACER_SIZE 100
132 static char bootup_tracer_buf[BOOTUP_TRACER_SIZE] __initdata;
133 static char *default_bootup_tracer;
135 static int __init set_ftrace(char *str)
137 strncpy(bootup_tracer_buf, str, BOOTUP_TRACER_SIZE);
138 default_bootup_tracer = bootup_tracer_buf;
139 /* We are using ftrace early, expand it */
140 ring_buffer_expanded = 1;
143 __setup("ftrace=", set_ftrace);
145 static int __init set_ftrace_dump_on_oops(char *str)
147 ftrace_dump_on_oops = 1;
150 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
152 unsigned long long ns2usecs(cycle_t nsec)
160 * The global_trace is the descriptor that holds the tracing
161 * buffers for the live tracing. For each CPU, it contains
162 * a link list of pages that will store trace entries. The
163 * page descriptor of the pages in the memory is used to hold
164 * the link list by linking the lru item in the page descriptor
165 * to each of the pages in the buffer per CPU.
167 * For each active CPU there is a data field that holds the
168 * pages for the buffer for that CPU. Each CPU has the same number
169 * of pages allocated for its buffer.
171 static struct trace_array global_trace;
173 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
175 int filter_current_check_discard(struct ftrace_event_call *call, void *rec,
176 struct ring_buffer_event *event)
178 return filter_check_discard(call, rec, global_trace.buffer, event);
180 EXPORT_SYMBOL_GPL(filter_current_check_discard);
182 cycle_t ftrace_now(int cpu)
186 /* Early boot up does not have a buffer yet */
187 if (!global_trace.buffer)
188 return trace_clock_local();
190 ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
191 ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
197 * The max_tr is used to snapshot the global_trace when a maximum
198 * latency is reached. Some tracers will use this to store a maximum
199 * trace while it continues examining live traces.
201 * The buffers for the max_tr are set up the same as the global_trace.
202 * When a snapshot is taken, the link list of the max_tr is swapped
203 * with the link list of the global_trace and the buffers are reset for
204 * the global_trace so the tracing can continue.
206 static struct trace_array max_tr;
208 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
210 /* tracer_enabled is used to toggle activation of a tracer */
211 static int tracer_enabled = 1;
214 * tracing_is_enabled - return tracer_enabled status
216 * This function is used by other tracers to know the status
217 * of the tracer_enabled flag. Tracers may use this function
218 * to know if it should enable their features when starting
219 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
221 int tracing_is_enabled(void)
223 return tracer_enabled;
227 * trace_buf_size is the size in bytes that is allocated
228 * for a buffer. Note, the number of bytes is always rounded
231 * This number is purposely set to a low number of 16384.
232 * If the dump on oops happens, it will be much appreciated
233 * to not have to wait for all that output. Anyway this can be
234 * boot time and run time configurable.
236 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
238 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
240 /* trace_types holds a link list of available tracers. */
241 static struct tracer *trace_types __read_mostly;
243 /* current_trace points to the tracer that is currently active */
244 static struct tracer *current_trace __read_mostly;
247 * max_tracer_type_len is used to simplify the allocating of
248 * buffers to read userspace tracer names. We keep track of
249 * the longest tracer name registered.
251 static int max_tracer_type_len;
254 * trace_types_lock is used to protect the trace_types list.
255 * This lock is also used to keep user access serialized.
256 * Accesses from userspace will grab this lock while userspace
257 * activities happen inside the kernel.
259 static DEFINE_MUTEX(trace_types_lock);
261 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
262 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
264 /* trace_flags holds trace_options default values */
265 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
266 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
267 TRACE_ITER_GRAPH_TIME;
270 * trace_wake_up - wake up tasks waiting for trace input
272 * Simply wakes up any task that is blocked on the trace_wait
273 * queue. These is used with trace_poll for tasks polling the trace.
275 void trace_wake_up(void)
278 * The runqueue_is_locked() can fail, but this is the best we
281 if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
282 wake_up(&trace_wait);
285 static int __init set_buf_size(char *str)
287 unsigned long buf_size;
291 buf_size = memparse(str, &str);
292 /* nr_entries can not be zero */
295 trace_buf_size = buf_size;
298 __setup("trace_buf_size=", set_buf_size);
300 unsigned long nsecs_to_usecs(unsigned long nsecs)
305 /* These must match the bit postions in trace_iterator_flags */
306 static const char *trace_options[] = {
333 * ftrace_max_lock is used to protect the swapping of buffers
334 * when taking a max snapshot. The buffers themselves are
335 * protected by per_cpu spinlocks. But the action of the swap
336 * needs its own lock.
338 * This is defined as a raw_spinlock_t in order to help
339 * with performance when lockdep debugging is enabled.
341 static raw_spinlock_t ftrace_max_lock =
342 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
345 * Copy the new maximum trace into the separate maximum-trace
346 * structure. (this way the maximum trace is permanently saved,
347 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
350 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
352 struct trace_array_cpu *data = tr->data[cpu];
355 max_tr.time_start = data->preempt_timestamp;
357 data = max_tr.data[cpu];
358 data->saved_latency = tracing_max_latency;
360 memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
361 data->pid = tsk->pid;
362 data->uid = task_uid(tsk);
363 data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
364 data->policy = tsk->policy;
365 data->rt_priority = tsk->rt_priority;
367 /* record this tasks comm */
368 tracing_record_cmdline(tsk);
371 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
379 if (s->len <= s->readpos)
382 len = s->len - s->readpos;
385 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
395 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
400 if (s->len <= s->readpos)
403 len = s->len - s->readpos;
406 ret = memcpy(buf, s->buffer + s->readpos, cnt);
415 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
417 * @tsk: the task with the latency
418 * @cpu: The cpu that initiated the trace.
420 * Flip the buffers between the @tr and the max_tr and record information
421 * about which task was the cause of this latency.
424 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
426 struct ring_buffer *buf = tr->buffer;
428 WARN_ON_ONCE(!irqs_disabled());
429 __raw_spin_lock(&ftrace_max_lock);
431 tr->buffer = max_tr.buffer;
434 ftrace_disable_cpu();
435 ring_buffer_reset(tr->buffer);
438 __update_max_tr(tr, tsk, cpu);
439 __raw_spin_unlock(&ftrace_max_lock);
443 * update_max_tr_single - only copy one trace over, and reset the rest
445 * @tsk - task with the latency
446 * @cpu - the cpu of the buffer to copy.
448 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
451 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
455 WARN_ON_ONCE(!irqs_disabled());
456 __raw_spin_lock(&ftrace_max_lock);
458 ftrace_disable_cpu();
460 ring_buffer_reset(max_tr.buffer);
461 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
465 WARN_ON_ONCE(ret && ret != -EAGAIN);
467 __update_max_tr(tr, tsk, cpu);
468 __raw_spin_unlock(&ftrace_max_lock);
472 * register_tracer - register a tracer with the ftrace system.
473 * @type - the plugin for the tracer
475 * Register a new plugin tracer.
477 int register_tracer(struct tracer *type)
478 __releases(kernel_lock)
479 __acquires(kernel_lock)
486 pr_info("Tracer must have a name\n");
491 * When this gets called we hold the BKL which means that
492 * preemption is disabled. Various trace selftests however
493 * need to disable and enable preemption for successful tests.
494 * So we drop the BKL here and grab it after the tests again.
497 mutex_lock(&trace_types_lock);
499 tracing_selftest_running = true;
501 for (t = trace_types; t; t = t->next) {
502 if (strcmp(type->name, t->name) == 0) {
504 pr_info("Trace %s already registered\n",
512 type->set_flag = &dummy_set_flag;
514 type->flags = &dummy_tracer_flags;
516 if (!type->flags->opts)
517 type->flags->opts = dummy_tracer_opt;
518 if (!type->wait_pipe)
519 type->wait_pipe = default_wait_pipe;
522 #ifdef CONFIG_FTRACE_STARTUP_TEST
523 if (type->selftest && !tracing_selftest_disabled) {
524 struct tracer *saved_tracer = current_trace;
525 struct trace_array *tr = &global_trace;
529 * Run a selftest on this tracer.
530 * Here we reset the trace buffer, and set the current
531 * tracer to be this tracer. The tracer can then run some
532 * internal tracing to verify that everything is in order.
533 * If we fail, we do not register this tracer.
535 for_each_tracing_cpu(i)
536 tracing_reset(tr, i);
538 current_trace = type;
539 /* the test is responsible for initializing and enabling */
540 pr_info("Testing tracer %s: ", type->name);
541 ret = type->selftest(type, tr);
542 /* the test is responsible for resetting too */
543 current_trace = saved_tracer;
545 printk(KERN_CONT "FAILED!\n");
548 /* Only reset on passing, to avoid touching corrupted buffers */
549 for_each_tracing_cpu(i)
550 tracing_reset(tr, i);
552 printk(KERN_CONT "PASSED\n");
556 type->next = trace_types;
558 len = strlen(type->name);
559 if (len > max_tracer_type_len)
560 max_tracer_type_len = len;
563 tracing_selftest_running = false;
564 mutex_unlock(&trace_types_lock);
566 if (ret || !default_bootup_tracer)
569 if (strncmp(default_bootup_tracer, type->name, BOOTUP_TRACER_SIZE))
572 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
573 /* Do we want this tracer to start on bootup? */
574 tracing_set_tracer(type->name);
575 default_bootup_tracer = NULL;
576 /* disable other selftests, since this will break it. */
577 tracing_selftest_disabled = 1;
578 #ifdef CONFIG_FTRACE_STARTUP_TEST
579 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
588 void unregister_tracer(struct tracer *type)
593 mutex_lock(&trace_types_lock);
594 for (t = &trace_types; *t; t = &(*t)->next) {
598 pr_info("Trace %s not registered\n", type->name);
604 if (type == current_trace && tracer_enabled) {
607 if (current_trace->stop)
608 current_trace->stop(&global_trace);
609 current_trace = &nop_trace;
612 if (strlen(type->name) != max_tracer_type_len)
615 max_tracer_type_len = 0;
616 for (t = &trace_types; *t; t = &(*t)->next) {
617 len = strlen((*t)->name);
618 if (len > max_tracer_type_len)
619 max_tracer_type_len = len;
622 mutex_unlock(&trace_types_lock);
625 void tracing_reset(struct trace_array *tr, int cpu)
627 ftrace_disable_cpu();
628 ring_buffer_reset_cpu(tr->buffer, cpu);
632 void tracing_reset_online_cpus(struct trace_array *tr)
636 tr->time_start = ftrace_now(tr->cpu);
638 for_each_online_cpu(cpu)
639 tracing_reset(tr, cpu);
642 void tracing_reset_current(int cpu)
644 tracing_reset(&global_trace, cpu);
647 void tracing_reset_current_online_cpus(void)
649 tracing_reset_online_cpus(&global_trace);
652 #define SAVED_CMDLINES 128
653 #define NO_CMDLINE_MAP UINT_MAX
654 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
655 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
656 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
657 static int cmdline_idx;
658 static raw_spinlock_t trace_cmdline_lock = __RAW_SPIN_LOCK_UNLOCKED;
660 /* temporary disable recording */
661 static atomic_t trace_record_cmdline_disabled __read_mostly;
663 static void trace_init_cmdlines(void)
665 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
666 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
670 static int trace_stop_count;
671 static DEFINE_SPINLOCK(tracing_start_lock);
674 * ftrace_off_permanent - disable all ftrace code permanently
676 * This should only be called when a serious anomally has
677 * been detected. This will turn off the function tracing,
678 * ring buffers, and other tracing utilites. It takes no
679 * locks and can be called from any context.
681 void ftrace_off_permanent(void)
683 tracing_disabled = 1;
685 tracing_off_permanent();
689 * tracing_start - quick start of the tracer
691 * If tracing is enabled but was stopped by tracing_stop,
692 * this will start the tracer back up.
694 void tracing_start(void)
696 struct ring_buffer *buffer;
699 if (tracing_disabled)
702 spin_lock_irqsave(&tracing_start_lock, flags);
703 if (--trace_stop_count) {
704 if (trace_stop_count < 0) {
705 /* Someone screwed up their debugging */
707 trace_stop_count = 0;
713 buffer = global_trace.buffer;
715 ring_buffer_record_enable(buffer);
717 buffer = max_tr.buffer;
719 ring_buffer_record_enable(buffer);
723 spin_unlock_irqrestore(&tracing_start_lock, flags);
727 * tracing_stop - quick stop of the tracer
729 * Light weight way to stop tracing. Use in conjunction with
732 void tracing_stop(void)
734 struct ring_buffer *buffer;
738 spin_lock_irqsave(&tracing_start_lock, flags);
739 if (trace_stop_count++)
742 buffer = global_trace.buffer;
744 ring_buffer_record_disable(buffer);
746 buffer = max_tr.buffer;
748 ring_buffer_record_disable(buffer);
751 spin_unlock_irqrestore(&tracing_start_lock, flags);
754 void trace_stop_cmdline_recording(void);
756 static void trace_save_cmdline(struct task_struct *tsk)
760 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
764 * It's not the end of the world if we don't get
765 * the lock, but we also don't want to spin
766 * nor do we want to disable interrupts,
767 * so if we miss here, then better luck next time.
769 if (!__raw_spin_trylock(&trace_cmdline_lock))
772 idx = map_pid_to_cmdline[tsk->pid];
773 if (idx == NO_CMDLINE_MAP) {
774 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
777 * Check whether the cmdline buffer at idx has a pid
778 * mapped. We are going to overwrite that entry so we
779 * need to clear the map_pid_to_cmdline. Otherwise we
780 * would read the new comm for the old pid.
782 pid = map_cmdline_to_pid[idx];
783 if (pid != NO_CMDLINE_MAP)
784 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
786 map_cmdline_to_pid[idx] = tsk->pid;
787 map_pid_to_cmdline[tsk->pid] = idx;
792 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
794 __raw_spin_unlock(&trace_cmdline_lock);
797 void trace_find_cmdline(int pid, char comm[])
802 strcpy(comm, "<idle>");
806 if (pid > PID_MAX_DEFAULT) {
807 strcpy(comm, "<...>");
812 __raw_spin_lock(&trace_cmdline_lock);
813 map = map_pid_to_cmdline[pid];
814 if (map != NO_CMDLINE_MAP)
815 strcpy(comm, saved_cmdlines[map]);
817 strcpy(comm, "<...>");
819 __raw_spin_unlock(&trace_cmdline_lock);
823 void tracing_record_cmdline(struct task_struct *tsk)
825 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
829 trace_save_cmdline(tsk);
833 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
836 struct task_struct *tsk = current;
838 entry->preempt_count = pc & 0xff;
839 entry->pid = (tsk) ? tsk->pid : 0;
840 entry->tgid = (tsk) ? tsk->tgid : 0;
842 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
843 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
845 TRACE_FLAG_IRQS_NOSUPPORT |
847 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
848 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
849 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
852 struct ring_buffer_event *trace_buffer_lock_reserve(struct trace_array *tr,
855 unsigned long flags, int pc)
857 struct ring_buffer_event *event;
859 event = ring_buffer_lock_reserve(tr->buffer, len);
861 struct trace_entry *ent = ring_buffer_event_data(event);
863 tracing_generic_entry_update(ent, flags, pc);
870 static inline void __trace_buffer_unlock_commit(struct trace_array *tr,
871 struct ring_buffer_event *event,
872 unsigned long flags, int pc,
875 ring_buffer_unlock_commit(tr->buffer, event);
877 ftrace_trace_stack(tr, flags, 6, pc);
878 ftrace_trace_userstack(tr, flags, pc);
884 void trace_buffer_unlock_commit(struct trace_array *tr,
885 struct ring_buffer_event *event,
886 unsigned long flags, int pc)
888 __trace_buffer_unlock_commit(tr, event, flags, pc, 1);
891 struct ring_buffer_event *
892 trace_current_buffer_lock_reserve(int type, unsigned long len,
893 unsigned long flags, int pc)
895 return trace_buffer_lock_reserve(&global_trace,
896 type, len, flags, pc);
898 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
900 void trace_current_buffer_unlock_commit(struct ring_buffer_event *event,
901 unsigned long flags, int pc)
903 __trace_buffer_unlock_commit(&global_trace, event, flags, pc, 1);
905 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
907 void trace_nowake_buffer_unlock_commit(struct ring_buffer_event *event,
908 unsigned long flags, int pc)
910 __trace_buffer_unlock_commit(&global_trace, event, flags, pc, 0);
912 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
914 void trace_current_buffer_discard_commit(struct ring_buffer_event *event)
916 ring_buffer_discard_commit(global_trace.buffer, event);
918 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
921 trace_function(struct trace_array *tr,
922 unsigned long ip, unsigned long parent_ip, unsigned long flags,
925 struct ftrace_event_call *call = &event_function;
926 struct ring_buffer_event *event;
927 struct ftrace_entry *entry;
929 /* If we are reading the ring buffer, don't trace */
930 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
933 event = trace_buffer_lock_reserve(tr, TRACE_FN, sizeof(*entry),
937 entry = ring_buffer_event_data(event);
939 entry->parent_ip = parent_ip;
941 if (!filter_check_discard(call, entry, tr->buffer, event))
942 ring_buffer_unlock_commit(tr->buffer, event);
945 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
946 static int __trace_graph_entry(struct trace_array *tr,
947 struct ftrace_graph_ent *trace,
951 struct ftrace_event_call *call = &event_funcgraph_entry;
952 struct ring_buffer_event *event;
953 struct ftrace_graph_ent_entry *entry;
955 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
958 event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_ENT,
959 sizeof(*entry), flags, pc);
962 entry = ring_buffer_event_data(event);
963 entry->graph_ent = *trace;
964 if (!filter_current_check_discard(call, entry, event))
965 ring_buffer_unlock_commit(global_trace.buffer, event);
970 static void __trace_graph_return(struct trace_array *tr,
971 struct ftrace_graph_ret *trace,
975 struct ftrace_event_call *call = &event_funcgraph_exit;
976 struct ring_buffer_event *event;
977 struct ftrace_graph_ret_entry *entry;
979 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
982 event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_RET,
983 sizeof(*entry), flags, pc);
986 entry = ring_buffer_event_data(event);
988 if (!filter_current_check_discard(call, entry, event))
989 ring_buffer_unlock_commit(global_trace.buffer, event);
994 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
995 unsigned long ip, unsigned long parent_ip, unsigned long flags,
998 if (likely(!atomic_read(&data->disabled)))
999 trace_function(tr, ip, parent_ip, flags, pc);
1002 #ifdef CONFIG_STACKTRACE
1003 static void __ftrace_trace_stack(struct trace_array *tr,
1004 unsigned long flags,
1007 struct ftrace_event_call *call = &event_kernel_stack;
1008 struct ring_buffer_event *event;
1009 struct stack_entry *entry;
1010 struct stack_trace trace;
1012 event = trace_buffer_lock_reserve(tr, TRACE_STACK,
1013 sizeof(*entry), flags, pc);
1016 entry = ring_buffer_event_data(event);
1017 memset(&entry->caller, 0, sizeof(entry->caller));
1019 trace.nr_entries = 0;
1020 trace.max_entries = FTRACE_STACK_ENTRIES;
1022 trace.entries = entry->caller;
1024 save_stack_trace(&trace);
1025 if (!filter_check_discard(call, entry, tr->buffer, event))
1026 ring_buffer_unlock_commit(tr->buffer, event);
1029 void ftrace_trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1032 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1035 __ftrace_trace_stack(tr, flags, skip, pc);
1038 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1041 __ftrace_trace_stack(tr, flags, skip, pc);
1044 void ftrace_trace_userstack(struct trace_array *tr, unsigned long flags, int pc)
1046 struct ftrace_event_call *call = &event_user_stack;
1047 struct ring_buffer_event *event;
1048 struct userstack_entry *entry;
1049 struct stack_trace trace;
1051 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1054 event = trace_buffer_lock_reserve(tr, TRACE_USER_STACK,
1055 sizeof(*entry), flags, pc);
1058 entry = ring_buffer_event_data(event);
1060 memset(&entry->caller, 0, sizeof(entry->caller));
1062 trace.nr_entries = 0;
1063 trace.max_entries = FTRACE_STACK_ENTRIES;
1065 trace.entries = entry->caller;
1067 save_stack_trace_user(&trace);
1068 if (!filter_check_discard(call, entry, tr->buffer, event))
1069 ring_buffer_unlock_commit(tr->buffer, event);
1073 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1075 ftrace_trace_userstack(tr, flags, preempt_count());
1079 #endif /* CONFIG_STACKTRACE */
1082 ftrace_trace_special(void *__tr,
1083 unsigned long arg1, unsigned long arg2, unsigned long arg3,
1086 struct ring_buffer_event *event;
1087 struct trace_array *tr = __tr;
1088 struct special_entry *entry;
1090 event = trace_buffer_lock_reserve(tr, TRACE_SPECIAL,
1091 sizeof(*entry), 0, pc);
1094 entry = ring_buffer_event_data(event);
1098 trace_buffer_unlock_commit(tr, event, 0, pc);
1102 __trace_special(void *__tr, void *__data,
1103 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1105 ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1109 tracing_sched_switch_trace(struct trace_array *tr,
1110 struct task_struct *prev,
1111 struct task_struct *next,
1112 unsigned long flags, int pc)
1114 struct ftrace_event_call *call = &event_context_switch;
1115 struct ring_buffer_event *event;
1116 struct ctx_switch_entry *entry;
1118 event = trace_buffer_lock_reserve(tr, TRACE_CTX,
1119 sizeof(*entry), flags, pc);
1122 entry = ring_buffer_event_data(event);
1123 entry->prev_pid = prev->pid;
1124 entry->prev_prio = prev->prio;
1125 entry->prev_state = prev->state;
1126 entry->next_pid = next->pid;
1127 entry->next_prio = next->prio;
1128 entry->next_state = next->state;
1129 entry->next_cpu = task_cpu(next);
1131 if (!filter_check_discard(call, entry, tr->buffer, event))
1132 trace_buffer_unlock_commit(tr, event, flags, pc);
1136 tracing_sched_wakeup_trace(struct trace_array *tr,
1137 struct task_struct *wakee,
1138 struct task_struct *curr,
1139 unsigned long flags, int pc)
1141 struct ftrace_event_call *call = &event_wakeup;
1142 struct ring_buffer_event *event;
1143 struct ctx_switch_entry *entry;
1145 event = trace_buffer_lock_reserve(tr, TRACE_WAKE,
1146 sizeof(*entry), flags, pc);
1149 entry = ring_buffer_event_data(event);
1150 entry->prev_pid = curr->pid;
1151 entry->prev_prio = curr->prio;
1152 entry->prev_state = curr->state;
1153 entry->next_pid = wakee->pid;
1154 entry->next_prio = wakee->prio;
1155 entry->next_state = wakee->state;
1156 entry->next_cpu = task_cpu(wakee);
1158 if (!filter_check_discard(call, entry, tr->buffer, event))
1159 ring_buffer_unlock_commit(tr->buffer, event);
1160 ftrace_trace_stack(tr, flags, 6, pc);
1161 ftrace_trace_userstack(tr, flags, pc);
1165 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1167 struct trace_array *tr = &global_trace;
1168 struct trace_array_cpu *data;
1169 unsigned long flags;
1173 if (tracing_disabled)
1176 pc = preempt_count();
1177 local_irq_save(flags);
1178 cpu = raw_smp_processor_id();
1179 data = tr->data[cpu];
1181 if (likely(atomic_inc_return(&data->disabled) == 1))
1182 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1184 atomic_dec(&data->disabled);
1185 local_irq_restore(flags);
1188 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1189 int trace_graph_entry(struct ftrace_graph_ent *trace)
1191 struct trace_array *tr = &global_trace;
1192 struct trace_array_cpu *data;
1193 unsigned long flags;
1199 if (!ftrace_trace_task(current))
1202 if (!ftrace_graph_addr(trace->func))
1205 local_irq_save(flags);
1206 cpu = raw_smp_processor_id();
1207 data = tr->data[cpu];
1208 disabled = atomic_inc_return(&data->disabled);
1209 if (likely(disabled == 1)) {
1210 pc = preempt_count();
1211 ret = __trace_graph_entry(tr, trace, flags, pc);
1215 /* Only do the atomic if it is not already set */
1216 if (!test_tsk_trace_graph(current))
1217 set_tsk_trace_graph(current);
1219 atomic_dec(&data->disabled);
1220 local_irq_restore(flags);
1225 void trace_graph_return(struct ftrace_graph_ret *trace)
1227 struct trace_array *tr = &global_trace;
1228 struct trace_array_cpu *data;
1229 unsigned long flags;
1234 local_irq_save(flags);
1235 cpu = raw_smp_processor_id();
1236 data = tr->data[cpu];
1237 disabled = atomic_inc_return(&data->disabled);
1238 if (likely(disabled == 1)) {
1239 pc = preempt_count();
1240 __trace_graph_return(tr, trace, flags, pc);
1243 clear_tsk_trace_graph(current);
1244 atomic_dec(&data->disabled);
1245 local_irq_restore(flags);
1247 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1251 * trace_vbprintk - write binary msg to tracing buffer
1254 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1256 static raw_spinlock_t trace_buf_lock =
1257 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
1258 static u32 trace_buf[TRACE_BUF_SIZE];
1260 struct ftrace_event_call *call = &event_bprint;
1261 struct ring_buffer_event *event;
1262 struct trace_array *tr = &global_trace;
1263 struct trace_array_cpu *data;
1264 struct bprint_entry *entry;
1265 unsigned long flags;
1268 int cpu, len = 0, size, pc;
1270 if (unlikely(tracing_selftest_running || tracing_disabled))
1273 /* Don't pollute graph traces with trace_vprintk internals */
1274 pause_graph_tracing();
1276 pc = preempt_count();
1277 resched = ftrace_preempt_disable();
1278 cpu = raw_smp_processor_id();
1279 data = tr->data[cpu];
1281 disable = atomic_inc_return(&data->disabled);
1282 if (unlikely(disable != 1))
1285 /* Lockdep uses trace_printk for lock tracing */
1286 local_irq_save(flags);
1287 __raw_spin_lock(&trace_buf_lock);
1288 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1290 if (len > TRACE_BUF_SIZE || len < 0)
1293 size = sizeof(*entry) + sizeof(u32) * len;
1294 event = trace_buffer_lock_reserve(tr, TRACE_BPRINT, size, flags, pc);
1297 entry = ring_buffer_event_data(event);
1301 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1302 if (!filter_check_discard(call, entry, tr->buffer, event))
1303 ring_buffer_unlock_commit(tr->buffer, event);
1306 __raw_spin_unlock(&trace_buf_lock);
1307 local_irq_restore(flags);
1310 atomic_dec_return(&data->disabled);
1311 ftrace_preempt_enable(resched);
1312 unpause_graph_tracing();
1316 EXPORT_SYMBOL_GPL(trace_vbprintk);
1318 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1320 static raw_spinlock_t trace_buf_lock = __RAW_SPIN_LOCK_UNLOCKED;
1321 static char trace_buf[TRACE_BUF_SIZE];
1323 struct ftrace_event_call *call = &event_print;
1324 struct ring_buffer_event *event;
1325 struct trace_array *tr = &global_trace;
1326 struct trace_array_cpu *data;
1327 int cpu, len = 0, size, pc;
1328 struct print_entry *entry;
1329 unsigned long irq_flags;
1332 if (tracing_disabled || tracing_selftest_running)
1335 pc = preempt_count();
1336 preempt_disable_notrace();
1337 cpu = raw_smp_processor_id();
1338 data = tr->data[cpu];
1340 disable = atomic_inc_return(&data->disabled);
1341 if (unlikely(disable != 1))
1344 pause_graph_tracing();
1345 raw_local_irq_save(irq_flags);
1346 __raw_spin_lock(&trace_buf_lock);
1347 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1349 len = min(len, TRACE_BUF_SIZE-1);
1352 size = sizeof(*entry) + len + 1;
1353 event = trace_buffer_lock_reserve(tr, TRACE_PRINT, size, irq_flags, pc);
1356 entry = ring_buffer_event_data(event);
1359 memcpy(&entry->buf, trace_buf, len);
1360 entry->buf[len] = 0;
1361 if (!filter_check_discard(call, entry, tr->buffer, event))
1362 ring_buffer_unlock_commit(tr->buffer, event);
1365 __raw_spin_unlock(&trace_buf_lock);
1366 raw_local_irq_restore(irq_flags);
1367 unpause_graph_tracing();
1369 atomic_dec_return(&data->disabled);
1370 preempt_enable_notrace();
1374 EXPORT_SYMBOL_GPL(trace_vprintk);
1376 enum trace_file_type {
1377 TRACE_FILE_LAT_FMT = 1,
1378 TRACE_FILE_ANNOTATE = 2,
1381 static void trace_iterator_increment(struct trace_iterator *iter)
1383 /* Don't allow ftrace to trace into the ring buffers */
1384 ftrace_disable_cpu();
1387 if (iter->buffer_iter[iter->cpu])
1388 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1390 ftrace_enable_cpu();
1393 static struct trace_entry *
1394 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1396 struct ring_buffer_event *event;
1397 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1399 /* Don't allow ftrace to trace into the ring buffers */
1400 ftrace_disable_cpu();
1403 event = ring_buffer_iter_peek(buf_iter, ts);
1405 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1407 ftrace_enable_cpu();
1409 return event ? ring_buffer_event_data(event) : NULL;
1412 static struct trace_entry *
1413 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1415 struct ring_buffer *buffer = iter->tr->buffer;
1416 struct trace_entry *ent, *next = NULL;
1417 int cpu_file = iter->cpu_file;
1418 u64 next_ts = 0, ts;
1423 * If we are in a per_cpu trace file, don't bother by iterating over
1424 * all cpu and peek directly.
1426 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1427 if (ring_buffer_empty_cpu(buffer, cpu_file))
1429 ent = peek_next_entry(iter, cpu_file, ent_ts);
1431 *ent_cpu = cpu_file;
1436 for_each_tracing_cpu(cpu) {
1438 if (ring_buffer_empty_cpu(buffer, cpu))
1441 ent = peek_next_entry(iter, cpu, &ts);
1444 * Pick the entry with the smallest timestamp:
1446 if (ent && (!next || ts < next_ts)) {
1454 *ent_cpu = next_cpu;
1462 /* Find the next real entry, without updating the iterator itself */
1463 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1464 int *ent_cpu, u64 *ent_ts)
1466 return __find_next_entry(iter, ent_cpu, ent_ts);
1469 /* Find the next real entry, and increment the iterator to the next entry */
1470 static void *find_next_entry_inc(struct trace_iterator *iter)
1472 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1475 trace_iterator_increment(iter);
1477 return iter->ent ? iter : NULL;
1480 static void trace_consume(struct trace_iterator *iter)
1482 /* Don't allow ftrace to trace into the ring buffers */
1483 ftrace_disable_cpu();
1484 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1485 ftrace_enable_cpu();
1488 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1490 struct trace_iterator *iter = m->private;
1496 /* can't go backwards */
1501 ent = find_next_entry_inc(iter);
1505 while (ent && iter->idx < i)
1506 ent = find_next_entry_inc(iter);
1514 * No necessary locking here. The worst thing which can
1515 * happen is loosing events consumed at the same time
1516 * by a trace_pipe reader.
1517 * Other than that, we don't risk to crash the ring buffer
1518 * because it serializes the readers.
1520 * The current tracer is copied to avoid a global locking
1523 static void *s_start(struct seq_file *m, loff_t *pos)
1525 struct trace_iterator *iter = m->private;
1526 static struct tracer *old_tracer;
1527 int cpu_file = iter->cpu_file;
1532 /* copy the tracer to avoid using a global lock all around */
1533 mutex_lock(&trace_types_lock);
1534 if (unlikely(old_tracer != current_trace && current_trace)) {
1535 old_tracer = current_trace;
1536 *iter->trace = *current_trace;
1538 mutex_unlock(&trace_types_lock);
1540 atomic_inc(&trace_record_cmdline_disabled);
1542 if (*pos != iter->pos) {
1547 ftrace_disable_cpu();
1549 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1550 for_each_tracing_cpu(cpu)
1551 ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1553 ring_buffer_iter_reset(iter->buffer_iter[cpu_file]);
1556 ftrace_enable_cpu();
1558 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1563 p = s_next(m, p, &l);
1566 trace_event_read_lock();
1570 static void s_stop(struct seq_file *m, void *p)
1572 atomic_dec(&trace_record_cmdline_disabled);
1573 trace_event_read_unlock();
1576 static void print_lat_help_header(struct seq_file *m)
1578 seq_puts(m, "# _------=> CPU# \n");
1579 seq_puts(m, "# / _-----=> irqs-off \n");
1580 seq_puts(m, "# | / _----=> need-resched \n");
1581 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1582 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1583 seq_puts(m, "# |||| / \n");
1584 seq_puts(m, "# ||||| delay \n");
1585 seq_puts(m, "# cmd pid ||||| time | caller \n");
1586 seq_puts(m, "# \\ / ||||| \\ | / \n");
1589 static void print_func_help_header(struct seq_file *m)
1591 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1592 seq_puts(m, "# | | | | |\n");
1597 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1599 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1600 struct trace_array *tr = iter->tr;
1601 struct trace_array_cpu *data = tr->data[tr->cpu];
1602 struct tracer *type = current_trace;
1603 unsigned long total;
1604 unsigned long entries;
1605 const char *name = "preemption";
1610 entries = ring_buffer_entries(iter->tr->buffer);
1612 ring_buffer_overruns(iter->tr->buffer);
1614 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1616 seq_puts(m, "# -----------------------------------"
1617 "---------------------------------\n");
1618 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1619 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1620 nsecs_to_usecs(data->saved_latency),
1624 #if defined(CONFIG_PREEMPT_NONE)
1626 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1628 #elif defined(CONFIG_PREEMPT)
1633 /* These are reserved for later use */
1636 seq_printf(m, " #P:%d)\n", num_online_cpus());
1640 seq_puts(m, "# -----------------\n");
1641 seq_printf(m, "# | task: %.16s-%d "
1642 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1643 data->comm, data->pid, data->uid, data->nice,
1644 data->policy, data->rt_priority);
1645 seq_puts(m, "# -----------------\n");
1647 if (data->critical_start) {
1648 seq_puts(m, "# => started at: ");
1649 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1650 trace_print_seq(m, &iter->seq);
1651 seq_puts(m, "\n# => ended at: ");
1652 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1653 trace_print_seq(m, &iter->seq);
1660 static void test_cpu_buff_start(struct trace_iterator *iter)
1662 struct trace_seq *s = &iter->seq;
1664 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1667 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1670 if (cpumask_test_cpu(iter->cpu, iter->started))
1673 cpumask_set_cpu(iter->cpu, iter->started);
1675 /* Don't print started cpu buffer for the first entry of the trace */
1677 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1681 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1683 struct trace_seq *s = &iter->seq;
1684 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1685 struct trace_entry *entry;
1686 struct trace_event *event;
1690 test_cpu_buff_start(iter);
1692 event = ftrace_find_event(entry->type);
1694 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1695 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1696 if (!trace_print_lat_context(iter))
1699 if (!trace_print_context(iter))
1705 return event->trace(iter, sym_flags);
1707 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1710 return TRACE_TYPE_HANDLED;
1712 return TRACE_TYPE_PARTIAL_LINE;
1715 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1717 struct trace_seq *s = &iter->seq;
1718 struct trace_entry *entry;
1719 struct trace_event *event;
1723 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1724 if (!trace_seq_printf(s, "%d %d %llu ",
1725 entry->pid, iter->cpu, iter->ts))
1729 event = ftrace_find_event(entry->type);
1731 return event->raw(iter, 0);
1733 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1736 return TRACE_TYPE_HANDLED;
1738 return TRACE_TYPE_PARTIAL_LINE;
1741 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1743 struct trace_seq *s = &iter->seq;
1744 unsigned char newline = '\n';
1745 struct trace_entry *entry;
1746 struct trace_event *event;
1750 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1751 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1752 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1753 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1756 event = ftrace_find_event(entry->type);
1758 enum print_line_t ret = event->hex(iter, 0);
1759 if (ret != TRACE_TYPE_HANDLED)
1763 SEQ_PUT_FIELD_RET(s, newline);
1765 return TRACE_TYPE_HANDLED;
1768 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1770 struct trace_seq *s = &iter->seq;
1771 struct trace_entry *entry;
1772 struct trace_event *event;
1776 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1777 SEQ_PUT_FIELD_RET(s, entry->pid);
1778 SEQ_PUT_FIELD_RET(s, iter->cpu);
1779 SEQ_PUT_FIELD_RET(s, iter->ts);
1782 event = ftrace_find_event(entry->type);
1783 return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1786 static int trace_empty(struct trace_iterator *iter)
1790 /* If we are looking at one CPU buffer, only check that one */
1791 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1792 cpu = iter->cpu_file;
1793 if (iter->buffer_iter[cpu]) {
1794 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1797 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1803 for_each_tracing_cpu(cpu) {
1804 if (iter->buffer_iter[cpu]) {
1805 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1808 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1816 /* Called with trace_event_read_lock() held. */
1817 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1819 enum print_line_t ret;
1821 if (iter->trace && iter->trace->print_line) {
1822 ret = iter->trace->print_line(iter);
1823 if (ret != TRACE_TYPE_UNHANDLED)
1827 if (iter->ent->type == TRACE_BPRINT &&
1828 trace_flags & TRACE_ITER_PRINTK &&
1829 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1830 return trace_print_bprintk_msg_only(iter);
1832 if (iter->ent->type == TRACE_PRINT &&
1833 trace_flags & TRACE_ITER_PRINTK &&
1834 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1835 return trace_print_printk_msg_only(iter);
1837 if (trace_flags & TRACE_ITER_BIN)
1838 return print_bin_fmt(iter);
1840 if (trace_flags & TRACE_ITER_HEX)
1841 return print_hex_fmt(iter);
1843 if (trace_flags & TRACE_ITER_RAW)
1844 return print_raw_fmt(iter);
1846 return print_trace_fmt(iter);
1849 static int s_show(struct seq_file *m, void *v)
1851 struct trace_iterator *iter = v;
1853 if (iter->ent == NULL) {
1855 seq_printf(m, "# tracer: %s\n", iter->trace->name);
1858 if (iter->trace && iter->trace->print_header)
1859 iter->trace->print_header(m);
1860 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1861 /* print nothing if the buffers are empty */
1862 if (trace_empty(iter))
1864 print_trace_header(m, iter);
1865 if (!(trace_flags & TRACE_ITER_VERBOSE))
1866 print_lat_help_header(m);
1868 if (!(trace_flags & TRACE_ITER_VERBOSE))
1869 print_func_help_header(m);
1872 print_trace_line(iter);
1873 trace_print_seq(m, &iter->seq);
1879 static struct seq_operations tracer_seq_ops = {
1886 static struct trace_iterator *
1887 __tracing_open(struct inode *inode, struct file *file)
1889 long cpu_file = (long) inode->i_private;
1890 void *fail_ret = ERR_PTR(-ENOMEM);
1891 struct trace_iterator *iter;
1895 if (tracing_disabled)
1896 return ERR_PTR(-ENODEV);
1898 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1900 return ERR_PTR(-ENOMEM);
1903 * We make a copy of the current tracer to avoid concurrent
1904 * changes on it while we are reading.
1906 mutex_lock(&trace_types_lock);
1907 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
1912 *iter->trace = *current_trace;
1914 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL))
1917 cpumask_clear(iter->started);
1919 if (current_trace && current_trace->print_max)
1922 iter->tr = &global_trace;
1924 mutex_init(&iter->mutex);
1925 iter->cpu_file = cpu_file;
1927 /* Notify the tracer early; before we stop tracing. */
1928 if (iter->trace && iter->trace->open)
1929 iter->trace->open(iter);
1931 /* Annotate start of buffers if we had overruns */
1932 if (ring_buffer_overruns(iter->tr->buffer))
1933 iter->iter_flags |= TRACE_FILE_ANNOTATE;
1935 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
1936 for_each_tracing_cpu(cpu) {
1938 iter->buffer_iter[cpu] =
1939 ring_buffer_read_start(iter->tr->buffer, cpu);
1942 cpu = iter->cpu_file;
1943 iter->buffer_iter[cpu] =
1944 ring_buffer_read_start(iter->tr->buffer, cpu);
1947 /* TODO stop tracer */
1948 ret = seq_open(file, &tracer_seq_ops);
1950 fail_ret = ERR_PTR(ret);
1954 m = file->private_data;
1957 /* stop the trace while dumping */
1960 mutex_unlock(&trace_types_lock);
1965 for_each_tracing_cpu(cpu) {
1966 if (iter->buffer_iter[cpu])
1967 ring_buffer_read_finish(iter->buffer_iter[cpu]);
1969 free_cpumask_var(iter->started);
1971 mutex_unlock(&trace_types_lock);
1978 int tracing_open_generic(struct inode *inode, struct file *filp)
1980 if (tracing_disabled)
1983 filp->private_data = inode->i_private;
1987 static int tracing_release(struct inode *inode, struct file *file)
1989 struct seq_file *m = (struct seq_file *)file->private_data;
1990 struct trace_iterator *iter;
1993 if (!(file->f_mode & FMODE_READ))
1998 mutex_lock(&trace_types_lock);
1999 for_each_tracing_cpu(cpu) {
2000 if (iter->buffer_iter[cpu])
2001 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2004 if (iter->trace && iter->trace->close)
2005 iter->trace->close(iter);
2007 /* reenable tracing if it was previously enabled */
2009 mutex_unlock(&trace_types_lock);
2011 seq_release(inode, file);
2012 mutex_destroy(&iter->mutex);
2013 free_cpumask_var(iter->started);
2019 static int tracing_open(struct inode *inode, struct file *file)
2021 struct trace_iterator *iter;
2024 /* If this file was open for write, then erase contents */
2025 if ((file->f_mode & FMODE_WRITE) &&
2026 !(file->f_flags & O_APPEND)) {
2027 long cpu = (long) inode->i_private;
2029 if (cpu == TRACE_PIPE_ALL_CPU)
2030 tracing_reset_online_cpus(&global_trace);
2032 tracing_reset(&global_trace, cpu);
2035 if (file->f_mode & FMODE_READ) {
2036 iter = __tracing_open(inode, file);
2038 ret = PTR_ERR(iter);
2039 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2040 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2046 t_next(struct seq_file *m, void *v, loff_t *pos)
2048 struct tracer *t = v;
2058 static void *t_start(struct seq_file *m, loff_t *pos)
2063 mutex_lock(&trace_types_lock);
2064 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2070 static void t_stop(struct seq_file *m, void *p)
2072 mutex_unlock(&trace_types_lock);
2075 static int t_show(struct seq_file *m, void *v)
2077 struct tracer *t = v;
2082 seq_printf(m, "%s", t->name);
2091 static struct seq_operations show_traces_seq_ops = {
2098 static int show_traces_open(struct inode *inode, struct file *file)
2100 if (tracing_disabled)
2103 return seq_open(file, &show_traces_seq_ops);
2107 tracing_write_stub(struct file *filp, const char __user *ubuf,
2108 size_t count, loff_t *ppos)
2113 static const struct file_operations tracing_fops = {
2114 .open = tracing_open,
2116 .write = tracing_write_stub,
2117 .llseek = seq_lseek,
2118 .release = tracing_release,
2121 static const struct file_operations show_traces_fops = {
2122 .open = show_traces_open,
2124 .release = seq_release,
2128 * Only trace on a CPU if the bitmask is set:
2130 static cpumask_var_t tracing_cpumask;
2133 * The tracer itself will not take this lock, but still we want
2134 * to provide a consistent cpumask to user-space:
2136 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2139 * Temporary storage for the character representation of the
2140 * CPU bitmask (and one more byte for the newline):
2142 static char mask_str[NR_CPUS + 1];
2145 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2146 size_t count, loff_t *ppos)
2150 mutex_lock(&tracing_cpumask_update_lock);
2152 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2153 if (count - len < 2) {
2157 len += sprintf(mask_str + len, "\n");
2158 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2161 mutex_unlock(&tracing_cpumask_update_lock);
2167 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2168 size_t count, loff_t *ppos)
2171 cpumask_var_t tracing_cpumask_new;
2173 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2176 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2180 mutex_lock(&tracing_cpumask_update_lock);
2182 local_irq_disable();
2183 __raw_spin_lock(&ftrace_max_lock);
2184 for_each_tracing_cpu(cpu) {
2186 * Increase/decrease the disabled counter if we are
2187 * about to flip a bit in the cpumask:
2189 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2190 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2191 atomic_inc(&global_trace.data[cpu]->disabled);
2193 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2194 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2195 atomic_dec(&global_trace.data[cpu]->disabled);
2198 __raw_spin_unlock(&ftrace_max_lock);
2201 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2203 mutex_unlock(&tracing_cpumask_update_lock);
2204 free_cpumask_var(tracing_cpumask_new);
2209 free_cpumask_var(tracing_cpumask_new);
2214 static const struct file_operations tracing_cpumask_fops = {
2215 .open = tracing_open_generic,
2216 .read = tracing_cpumask_read,
2217 .write = tracing_cpumask_write,
2221 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2222 size_t cnt, loff_t *ppos)
2224 struct tracer_opt *trace_opts;
2232 /* calculate max size */
2233 for (i = 0; trace_options[i]; i++) {
2234 len += strlen(trace_options[i]);
2235 len += 3; /* "no" and newline */
2238 mutex_lock(&trace_types_lock);
2239 tracer_flags = current_trace->flags->val;
2240 trace_opts = current_trace->flags->opts;
2243 * Increase the size with names of options specific
2244 * of the current tracer.
2246 for (i = 0; trace_opts[i].name; i++) {
2247 len += strlen(trace_opts[i].name);
2248 len += 3; /* "no" and newline */
2252 buf = kmalloc(len + 1, GFP_KERNEL);
2254 mutex_unlock(&trace_types_lock);
2258 for (i = 0; trace_options[i]; i++) {
2259 if (trace_flags & (1 << i))
2260 r += sprintf(buf + r, "%s\n", trace_options[i]);
2262 r += sprintf(buf + r, "no%s\n", trace_options[i]);
2265 for (i = 0; trace_opts[i].name; i++) {
2266 if (tracer_flags & trace_opts[i].bit)
2267 r += sprintf(buf + r, "%s\n",
2268 trace_opts[i].name);
2270 r += sprintf(buf + r, "no%s\n",
2271 trace_opts[i].name);
2273 mutex_unlock(&trace_types_lock);
2275 WARN_ON(r >= len + 1);
2277 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2283 /* Try to assign a tracer specific option */
2284 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2286 struct tracer_flags *trace_flags = trace->flags;
2287 struct tracer_opt *opts = NULL;
2291 for (i = 0; trace_flags->opts[i].name; i++) {
2292 opts = &trace_flags->opts[i];
2293 len = strlen(opts->name);
2295 if (strncmp(cmp, opts->name, len) == 0) {
2296 ret = trace->set_flag(trace_flags->val,
2302 if (!trace_flags->opts[i].name)
2305 /* Refused to handle */
2310 trace_flags->val &= ~opts->bit;
2312 trace_flags->val |= opts->bit;
2317 static void set_tracer_flags(unsigned int mask, int enabled)
2319 /* do nothing if flag is already set */
2320 if (!!(trace_flags & mask) == !!enabled)
2324 trace_flags |= mask;
2326 trace_flags &= ~mask;
2328 if (mask == TRACE_ITER_GLOBAL_CLK) {
2332 func = trace_clock_global;
2334 func = trace_clock_local;
2336 mutex_lock(&trace_types_lock);
2337 ring_buffer_set_clock(global_trace.buffer, func);
2340 ring_buffer_set_clock(max_tr.buffer, func);
2341 mutex_unlock(&trace_types_lock);
2346 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2347 size_t cnt, loff_t *ppos)
2355 if (cnt >= sizeof(buf))
2358 if (copy_from_user(&buf, ubuf, cnt))
2363 if (strncmp(buf, "no", 2) == 0) {
2368 for (i = 0; trace_options[i]; i++) {
2369 int len = strlen(trace_options[i]);
2371 if (strncmp(cmp, trace_options[i], len) == 0) {
2372 set_tracer_flags(1 << i, !neg);
2377 /* If no option could be set, test the specific tracer options */
2378 if (!trace_options[i]) {
2379 mutex_lock(&trace_types_lock);
2380 ret = set_tracer_option(current_trace, cmp, neg);
2381 mutex_unlock(&trace_types_lock);
2391 static const struct file_operations tracing_iter_fops = {
2392 .open = tracing_open_generic,
2393 .read = tracing_trace_options_read,
2394 .write = tracing_trace_options_write,
2397 static const char readme_msg[] =
2398 "tracing mini-HOWTO:\n\n"
2399 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2400 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2401 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2402 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2404 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2405 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2407 "# cat /sys/kernel/debug/tracing/trace_options\n"
2408 "noprint-parent nosym-offset nosym-addr noverbose\n"
2409 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2410 "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2411 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2412 "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2416 tracing_readme_read(struct file *filp, char __user *ubuf,
2417 size_t cnt, loff_t *ppos)
2419 return simple_read_from_buffer(ubuf, cnt, ppos,
2420 readme_msg, strlen(readme_msg));
2423 static const struct file_operations tracing_readme_fops = {
2424 .open = tracing_open_generic,
2425 .read = tracing_readme_read,
2429 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2430 size_t cnt, loff_t *ppos)
2439 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2443 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2451 for (i = 0; i < SAVED_CMDLINES; i++) {
2454 pid = map_cmdline_to_pid[i];
2455 if (pid == -1 || pid == NO_CMDLINE_MAP)
2458 trace_find_cmdline(pid, buf_comm);
2459 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2464 len = simple_read_from_buffer(ubuf, cnt, ppos,
2473 static const struct file_operations tracing_saved_cmdlines_fops = {
2474 .open = tracing_open_generic,
2475 .read = tracing_saved_cmdlines_read,
2479 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2480 size_t cnt, loff_t *ppos)
2485 r = sprintf(buf, "%u\n", tracer_enabled);
2486 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2490 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2491 size_t cnt, loff_t *ppos)
2493 struct trace_array *tr = filp->private_data;
2498 if (cnt >= sizeof(buf))
2501 if (copy_from_user(&buf, ubuf, cnt))
2506 ret = strict_strtoul(buf, 10, &val);
2512 mutex_lock(&trace_types_lock);
2513 if (tracer_enabled ^ val) {
2516 if (current_trace->start)
2517 current_trace->start(tr);
2522 if (current_trace->stop)
2523 current_trace->stop(tr);
2526 mutex_unlock(&trace_types_lock);
2534 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2535 size_t cnt, loff_t *ppos)
2537 char buf[max_tracer_type_len+2];
2540 mutex_lock(&trace_types_lock);
2542 r = sprintf(buf, "%s\n", current_trace->name);
2544 r = sprintf(buf, "\n");
2545 mutex_unlock(&trace_types_lock);
2547 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2550 int tracer_init(struct tracer *t, struct trace_array *tr)
2552 tracing_reset_online_cpus(tr);
2556 static int tracing_resize_ring_buffer(unsigned long size)
2561 * If kernel or user changes the size of the ring buffer
2562 * we use the size that was given, and we can forget about
2563 * expanding it later.
2565 ring_buffer_expanded = 1;
2567 ret = ring_buffer_resize(global_trace.buffer, size);
2571 ret = ring_buffer_resize(max_tr.buffer, size);
2575 r = ring_buffer_resize(global_trace.buffer,
2576 global_trace.entries);
2579 * AARGH! We are left with different
2580 * size max buffer!!!!
2581 * The max buffer is our "snapshot" buffer.
2582 * When a tracer needs a snapshot (one of the
2583 * latency tracers), it swaps the max buffer
2584 * with the saved snap shot. We succeeded to
2585 * update the size of the main buffer, but failed to
2586 * update the size of the max buffer. But when we tried
2587 * to reset the main buffer to the original size, we
2588 * failed there too. This is very unlikely to
2589 * happen, but if it does, warn and kill all
2593 tracing_disabled = 1;
2598 global_trace.entries = size;
2604 * tracing_update_buffers - used by tracing facility to expand ring buffers
2606 * To save on memory when the tracing is never used on a system with it
2607 * configured in. The ring buffers are set to a minimum size. But once
2608 * a user starts to use the tracing facility, then they need to grow
2609 * to their default size.
2611 * This function is to be called when a tracer is about to be used.
2613 int tracing_update_buffers(void)
2617 mutex_lock(&trace_types_lock);
2618 if (!ring_buffer_expanded)
2619 ret = tracing_resize_ring_buffer(trace_buf_size);
2620 mutex_unlock(&trace_types_lock);
2625 struct trace_option_dentry;
2627 static struct trace_option_dentry *
2628 create_trace_option_files(struct tracer *tracer);
2631 destroy_trace_option_files(struct trace_option_dentry *topts);
2633 static int tracing_set_tracer(const char *buf)
2635 static struct trace_option_dentry *topts;
2636 struct trace_array *tr = &global_trace;
2640 mutex_lock(&trace_types_lock);
2642 if (!ring_buffer_expanded) {
2643 ret = tracing_resize_ring_buffer(trace_buf_size);
2649 for (t = trace_types; t; t = t->next) {
2650 if (strcmp(t->name, buf) == 0)
2657 if (t == current_trace)
2660 trace_branch_disable();
2661 if (current_trace && current_trace->reset)
2662 current_trace->reset(tr);
2664 destroy_trace_option_files(topts);
2668 topts = create_trace_option_files(current_trace);
2671 ret = tracer_init(t, tr);
2676 trace_branch_enable(tr);
2678 mutex_unlock(&trace_types_lock);
2684 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2685 size_t cnt, loff_t *ppos)
2687 char buf[max_tracer_type_len+1];
2694 if (cnt > max_tracer_type_len)
2695 cnt = max_tracer_type_len;
2697 if (copy_from_user(&buf, ubuf, cnt))
2702 /* strip ending whitespace. */
2703 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2706 err = tracing_set_tracer(buf);
2716 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2717 size_t cnt, loff_t *ppos)
2719 unsigned long *ptr = filp->private_data;
2723 r = snprintf(buf, sizeof(buf), "%ld\n",
2724 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2725 if (r > sizeof(buf))
2727 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2731 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2732 size_t cnt, loff_t *ppos)
2734 unsigned long *ptr = filp->private_data;
2739 if (cnt >= sizeof(buf))
2742 if (copy_from_user(&buf, ubuf, cnt))
2747 ret = strict_strtoul(buf, 10, &val);
2756 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2758 long cpu_file = (long) inode->i_private;
2759 struct trace_iterator *iter;
2762 if (tracing_disabled)
2765 mutex_lock(&trace_types_lock);
2767 /* We only allow one reader per cpu */
2768 if (cpu_file == TRACE_PIPE_ALL_CPU) {
2769 if (!cpumask_empty(tracing_reader_cpumask)) {
2773 cpumask_setall(tracing_reader_cpumask);
2775 if (!cpumask_test_cpu(cpu_file, tracing_reader_cpumask))
2776 cpumask_set_cpu(cpu_file, tracing_reader_cpumask);
2783 /* create a buffer to store the information to pass to userspace */
2784 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2791 * We make a copy of the current tracer to avoid concurrent
2792 * changes on it while we are reading.
2794 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2800 *iter->trace = *current_trace;
2802 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2807 /* trace pipe does not show start of buffer */
2808 cpumask_setall(iter->started);
2810 if (trace_flags & TRACE_ITER_LATENCY_FMT)
2811 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2813 iter->cpu_file = cpu_file;
2814 iter->tr = &global_trace;
2815 mutex_init(&iter->mutex);
2816 filp->private_data = iter;
2818 if (iter->trace->pipe_open)
2819 iter->trace->pipe_open(iter);
2822 mutex_unlock(&trace_types_lock);
2828 mutex_unlock(&trace_types_lock);
2832 static int tracing_release_pipe(struct inode *inode, struct file *file)
2834 struct trace_iterator *iter = file->private_data;
2836 mutex_lock(&trace_types_lock);
2838 if (iter->cpu_file == TRACE_PIPE_ALL_CPU)
2839 cpumask_clear(tracing_reader_cpumask);
2841 cpumask_clear_cpu(iter->cpu_file, tracing_reader_cpumask);
2843 mutex_unlock(&trace_types_lock);
2845 free_cpumask_var(iter->started);
2846 mutex_destroy(&iter->mutex);
2854 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2856 struct trace_iterator *iter = filp->private_data;
2858 if (trace_flags & TRACE_ITER_BLOCK) {
2860 * Always select as readable when in blocking mode
2862 return POLLIN | POLLRDNORM;
2864 if (!trace_empty(iter))
2865 return POLLIN | POLLRDNORM;
2866 poll_wait(filp, &trace_wait, poll_table);
2867 if (!trace_empty(iter))
2868 return POLLIN | POLLRDNORM;
2875 void default_wait_pipe(struct trace_iterator *iter)
2879 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
2881 if (trace_empty(iter))
2884 finish_wait(&trace_wait, &wait);
2888 * This is a make-shift waitqueue.
2889 * A tracer might use this callback on some rare cases:
2891 * 1) the current tracer might hold the runqueue lock when it wakes up
2892 * a reader, hence a deadlock (sched, function, and function graph tracers)
2893 * 2) the function tracers, trace all functions, we don't want
2894 * the overhead of calling wake_up and friends
2895 * (and tracing them too)
2897 * Anyway, this is really very primitive wakeup.
2899 void poll_wait_pipe(struct trace_iterator *iter)
2901 set_current_state(TASK_INTERRUPTIBLE);
2902 /* sleep for 100 msecs, and try again. */
2903 schedule_timeout(HZ / 10);
2906 /* Must be called with trace_types_lock mutex held. */
2907 static int tracing_wait_pipe(struct file *filp)
2909 struct trace_iterator *iter = filp->private_data;
2911 while (trace_empty(iter)) {
2913 if ((filp->f_flags & O_NONBLOCK)) {
2917 mutex_unlock(&iter->mutex);
2919 iter->trace->wait_pipe(iter);
2921 mutex_lock(&iter->mutex);
2923 if (signal_pending(current))
2927 * We block until we read something and tracing is disabled.
2928 * We still block if tracing is disabled, but we have never
2929 * read anything. This allows a user to cat this file, and
2930 * then enable tracing. But after we have read something,
2931 * we give an EOF when tracing is again disabled.
2933 * iter->pos will be 0 if we haven't read anything.
2935 if (!tracer_enabled && iter->pos)
2946 tracing_read_pipe(struct file *filp, char __user *ubuf,
2947 size_t cnt, loff_t *ppos)
2949 struct trace_iterator *iter = filp->private_data;
2950 static struct tracer *old_tracer;
2953 /* return any leftover data */
2954 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2958 trace_seq_init(&iter->seq);
2960 /* copy the tracer to avoid using a global lock all around */
2961 mutex_lock(&trace_types_lock);
2962 if (unlikely(old_tracer != current_trace && current_trace)) {
2963 old_tracer = current_trace;
2964 *iter->trace = *current_trace;
2966 mutex_unlock(&trace_types_lock);
2969 * Avoid more than one consumer on a single file descriptor
2970 * This is just a matter of traces coherency, the ring buffer itself
2973 mutex_lock(&iter->mutex);
2974 if (iter->trace->read) {
2975 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2981 sret = tracing_wait_pipe(filp);
2985 /* stop when tracing is finished */
2986 if (trace_empty(iter)) {
2991 if (cnt >= PAGE_SIZE)
2992 cnt = PAGE_SIZE - 1;
2994 /* reset all but tr, trace, and overruns */
2995 memset(&iter->seq, 0,
2996 sizeof(struct trace_iterator) -
2997 offsetof(struct trace_iterator, seq));
3000 trace_event_read_lock();
3001 while (find_next_entry_inc(iter) != NULL) {
3002 enum print_line_t ret;
3003 int len = iter->seq.len;
3005 ret = print_trace_line(iter);
3006 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3007 /* don't print partial lines */
3008 iter->seq.len = len;
3011 if (ret != TRACE_TYPE_NO_CONSUME)
3012 trace_consume(iter);
3014 if (iter->seq.len >= cnt)
3017 trace_event_read_unlock();
3019 /* Now copy what we have to the user */
3020 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3021 if (iter->seq.readpos >= iter->seq.len)
3022 trace_seq_init(&iter->seq);
3025 * If there was nothing to send to user, inspite of consuming trace
3026 * entries, go back to wait for more entries.
3032 mutex_unlock(&iter->mutex);
3037 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3038 struct pipe_buffer *buf)
3040 __free_page(buf->page);
3043 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3046 __free_page(spd->pages[idx]);
3049 static struct pipe_buf_operations tracing_pipe_buf_ops = {
3051 .map = generic_pipe_buf_map,
3052 .unmap = generic_pipe_buf_unmap,
3053 .confirm = generic_pipe_buf_confirm,
3054 .release = tracing_pipe_buf_release,
3055 .steal = generic_pipe_buf_steal,
3056 .get = generic_pipe_buf_get,
3060 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3065 /* Seq buffer is page-sized, exactly what we need. */
3067 count = iter->seq.len;
3068 ret = print_trace_line(iter);
3069 count = iter->seq.len - count;
3072 iter->seq.len -= count;
3075 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3076 iter->seq.len -= count;
3080 trace_consume(iter);
3082 if (!find_next_entry_inc(iter)) {
3092 static ssize_t tracing_splice_read_pipe(struct file *filp,
3094 struct pipe_inode_info *pipe,
3098 struct page *pages[PIPE_BUFFERS];
3099 struct partial_page partial[PIPE_BUFFERS];
3100 struct trace_iterator *iter = filp->private_data;
3101 struct splice_pipe_desc spd = {
3104 .nr_pages = 0, /* This gets updated below. */
3106 .ops = &tracing_pipe_buf_ops,
3107 .spd_release = tracing_spd_release_pipe,
3109 static struct tracer *old_tracer;
3114 /* copy the tracer to avoid using a global lock all around */
3115 mutex_lock(&trace_types_lock);
3116 if (unlikely(old_tracer != current_trace && current_trace)) {
3117 old_tracer = current_trace;
3118 *iter->trace = *current_trace;
3120 mutex_unlock(&trace_types_lock);
3122 mutex_lock(&iter->mutex);
3124 if (iter->trace->splice_read) {
3125 ret = iter->trace->splice_read(iter, filp,
3126 ppos, pipe, len, flags);
3131 ret = tracing_wait_pipe(filp);
3135 if (!iter->ent && !find_next_entry_inc(iter)) {
3140 trace_event_read_lock();
3142 /* Fill as many pages as possible. */
3143 for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
3144 pages[i] = alloc_page(GFP_KERNEL);
3148 rem = tracing_fill_pipe_page(rem, iter);
3150 /* Copy the data into the page, so we can start over. */
3151 ret = trace_seq_to_buffer(&iter->seq,
3152 page_address(pages[i]),
3155 __free_page(pages[i]);
3158 partial[i].offset = 0;
3159 partial[i].len = iter->seq.len;
3161 trace_seq_init(&iter->seq);
3164 trace_event_read_unlock();
3165 mutex_unlock(&iter->mutex);
3169 return splice_to_pipe(pipe, &spd);
3172 mutex_unlock(&iter->mutex);
3178 tracing_entries_read(struct file *filp, char __user *ubuf,
3179 size_t cnt, loff_t *ppos)
3181 struct trace_array *tr = filp->private_data;
3185 mutex_lock(&trace_types_lock);
3186 if (!ring_buffer_expanded)
3187 r = sprintf(buf, "%lu (expanded: %lu)\n",
3189 trace_buf_size >> 10);
3191 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3192 mutex_unlock(&trace_types_lock);
3194 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3198 tracing_entries_write(struct file *filp, const char __user *ubuf,
3199 size_t cnt, loff_t *ppos)
3205 if (cnt >= sizeof(buf))
3208 if (copy_from_user(&buf, ubuf, cnt))
3213 ret = strict_strtoul(buf, 10, &val);
3217 /* must have at least 1 entry */
3221 mutex_lock(&trace_types_lock);
3225 /* disable all cpu buffers */
3226 for_each_tracing_cpu(cpu) {
3227 if (global_trace.data[cpu])
3228 atomic_inc(&global_trace.data[cpu]->disabled);
3229 if (max_tr.data[cpu])
3230 atomic_inc(&max_tr.data[cpu]->disabled);
3233 /* value is in KB */
3236 if (val != global_trace.entries) {
3237 ret = tracing_resize_ring_buffer(val);
3246 /* If check pages failed, return ENOMEM */
3247 if (tracing_disabled)
3250 for_each_tracing_cpu(cpu) {
3251 if (global_trace.data[cpu])
3252 atomic_dec(&global_trace.data[cpu]->disabled);
3253 if (max_tr.data[cpu])
3254 atomic_dec(&max_tr.data[cpu]->disabled);
3258 max_tr.entries = global_trace.entries;
3259 mutex_unlock(&trace_types_lock);
3264 static int mark_printk(const char *fmt, ...)
3268 va_start(args, fmt);
3269 ret = trace_vprintk(0, fmt, args);
3275 tracing_mark_write(struct file *filp, const char __user *ubuf,
3276 size_t cnt, loff_t *fpos)
3281 if (tracing_disabled)
3284 if (cnt > TRACE_BUF_SIZE)
3285 cnt = TRACE_BUF_SIZE;
3287 buf = kmalloc(cnt + 1, GFP_KERNEL);
3291 if (copy_from_user(buf, ubuf, cnt)) {
3296 /* Cut from the first nil or newline. */
3298 end = strchr(buf, '\n');
3302 cnt = mark_printk("%s\n", buf);
3309 static const struct file_operations tracing_max_lat_fops = {
3310 .open = tracing_open_generic,
3311 .read = tracing_max_lat_read,
3312 .write = tracing_max_lat_write,
3315 static const struct file_operations tracing_ctrl_fops = {
3316 .open = tracing_open_generic,
3317 .read = tracing_ctrl_read,
3318 .write = tracing_ctrl_write,
3321 static const struct file_operations set_tracer_fops = {
3322 .open = tracing_open_generic,
3323 .read = tracing_set_trace_read,
3324 .write = tracing_set_trace_write,
3327 static const struct file_operations tracing_pipe_fops = {
3328 .open = tracing_open_pipe,
3329 .poll = tracing_poll_pipe,
3330 .read = tracing_read_pipe,
3331 .splice_read = tracing_splice_read_pipe,
3332 .release = tracing_release_pipe,
3335 static const struct file_operations tracing_entries_fops = {
3336 .open = tracing_open_generic,
3337 .read = tracing_entries_read,
3338 .write = tracing_entries_write,
3341 static const struct file_operations tracing_mark_fops = {
3342 .open = tracing_open_generic,
3343 .write = tracing_mark_write,
3346 struct ftrace_buffer_info {
3347 struct trace_array *tr;
3353 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3355 int cpu = (int)(long)inode->i_private;
3356 struct ftrace_buffer_info *info;
3358 if (tracing_disabled)
3361 info = kzalloc(sizeof(*info), GFP_KERNEL);
3365 info->tr = &global_trace;
3368 /* Force reading ring buffer for first read */
3369 info->read = (unsigned int)-1;
3371 filp->private_data = info;
3373 return nonseekable_open(inode, filp);
3377 tracing_buffers_read(struct file *filp, char __user *ubuf,
3378 size_t count, loff_t *ppos)
3380 struct ftrace_buffer_info *info = filp->private_data;
3389 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3393 /* Do we have previous read data to read? */
3394 if (info->read < PAGE_SIZE)
3399 ret = ring_buffer_read_page(info->tr->buffer,
3406 pos = ring_buffer_page_len(info->spare);
3408 if (pos < PAGE_SIZE)
3409 memset(info->spare + pos, 0, PAGE_SIZE - pos);
3412 size = PAGE_SIZE - info->read;
3416 ret = copy_to_user(ubuf, info->spare + info->read, size);
3427 static int tracing_buffers_release(struct inode *inode, struct file *file)
3429 struct ftrace_buffer_info *info = file->private_data;
3432 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3439 struct ring_buffer *buffer;
3444 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3445 struct pipe_buffer *buf)
3447 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3452 ring_buffer_free_read_page(ref->buffer, ref->page);
3457 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3458 struct pipe_buffer *buf)
3463 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3464 struct pipe_buffer *buf)
3466 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3471 /* Pipe buffer operations for a buffer. */
3472 static struct pipe_buf_operations buffer_pipe_buf_ops = {
3474 .map = generic_pipe_buf_map,
3475 .unmap = generic_pipe_buf_unmap,
3476 .confirm = generic_pipe_buf_confirm,
3477 .release = buffer_pipe_buf_release,
3478 .steal = buffer_pipe_buf_steal,
3479 .get = buffer_pipe_buf_get,
3483 * Callback from splice_to_pipe(), if we need to release some pages
3484 * at the end of the spd in case we error'ed out in filling the pipe.
3486 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3488 struct buffer_ref *ref =
3489 (struct buffer_ref *)spd->partial[i].private;
3494 ring_buffer_free_read_page(ref->buffer, ref->page);
3496 spd->partial[i].private = 0;
3500 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3501 struct pipe_inode_info *pipe, size_t len,
3504 struct ftrace_buffer_info *info = file->private_data;
3505 struct partial_page partial[PIPE_BUFFERS];
3506 struct page *pages[PIPE_BUFFERS];
3507 struct splice_pipe_desc spd = {
3511 .ops = &buffer_pipe_buf_ops,
3512 .spd_release = buffer_spd_release,
3514 struct buffer_ref *ref;
3515 int entries, size, i;
3518 if (*ppos & (PAGE_SIZE - 1)) {
3519 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3523 if (len & (PAGE_SIZE - 1)) {
3524 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3525 if (len < PAGE_SIZE)
3530 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3532 for (i = 0; i < PIPE_BUFFERS && len && entries; i++, len -= PAGE_SIZE) {
3536 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3541 ref->buffer = info->tr->buffer;
3542 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3548 r = ring_buffer_read_page(ref->buffer, &ref->page,
3551 ring_buffer_free_read_page(ref->buffer,
3558 * zero out any left over data, this is going to
3561 size = ring_buffer_page_len(ref->page);
3562 if (size < PAGE_SIZE)
3563 memset(ref->page + size, 0, PAGE_SIZE - size);
3565 page = virt_to_page(ref->page);
3567 spd.pages[i] = page;
3568 spd.partial[i].len = PAGE_SIZE;
3569 spd.partial[i].offset = 0;
3570 spd.partial[i].private = (unsigned long)ref;
3574 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3579 /* did we read anything? */
3580 if (!spd.nr_pages) {
3581 if (flags & SPLICE_F_NONBLOCK)
3589 ret = splice_to_pipe(pipe, &spd);
3594 static const struct file_operations tracing_buffers_fops = {
3595 .open = tracing_buffers_open,
3596 .read = tracing_buffers_read,
3597 .release = tracing_buffers_release,
3598 .splice_read = tracing_buffers_splice_read,
3599 .llseek = no_llseek,
3603 tracing_stats_read(struct file *filp, char __user *ubuf,
3604 size_t count, loff_t *ppos)
3606 unsigned long cpu = (unsigned long)filp->private_data;
3607 struct trace_array *tr = &global_trace;
3608 struct trace_seq *s;
3611 s = kmalloc(sizeof(*s), GFP_KERNEL);
3617 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3618 trace_seq_printf(s, "entries: %ld\n", cnt);
3620 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3621 trace_seq_printf(s, "overrun: %ld\n", cnt);
3623 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3624 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3626 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3633 static const struct file_operations tracing_stats_fops = {
3634 .open = tracing_open_generic,
3635 .read = tracing_stats_read,
3638 #ifdef CONFIG_DYNAMIC_FTRACE
3640 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3646 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3647 size_t cnt, loff_t *ppos)
3649 static char ftrace_dyn_info_buffer[1024];
3650 static DEFINE_MUTEX(dyn_info_mutex);
3651 unsigned long *p = filp->private_data;
3652 char *buf = ftrace_dyn_info_buffer;
3653 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3656 mutex_lock(&dyn_info_mutex);
3657 r = sprintf(buf, "%ld ", *p);
3659 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3662 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3664 mutex_unlock(&dyn_info_mutex);
3669 static const struct file_operations tracing_dyn_info_fops = {
3670 .open = tracing_open_generic,
3671 .read = tracing_read_dyn_info,
3675 static struct dentry *d_tracer;
3677 struct dentry *tracing_init_dentry(void)
3684 if (!debugfs_initialized())
3687 d_tracer = debugfs_create_dir("tracing", NULL);
3689 if (!d_tracer && !once) {
3691 pr_warning("Could not create debugfs directory 'tracing'\n");
3698 static struct dentry *d_percpu;
3700 struct dentry *tracing_dentry_percpu(void)
3703 struct dentry *d_tracer;
3708 d_tracer = tracing_init_dentry();
3713 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3715 if (!d_percpu && !once) {
3717 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3724 static void tracing_init_debugfs_percpu(long cpu)
3726 struct dentry *d_percpu = tracing_dentry_percpu();
3727 struct dentry *d_cpu;
3728 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3731 if (cpu > 999 || cpu < 0)
3734 sprintf(cpu_dir, "cpu%ld", cpu);
3735 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3737 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3741 /* per cpu trace_pipe */
3742 trace_create_file("trace_pipe", 0444, d_cpu,
3743 (void *) cpu, &tracing_pipe_fops);
3746 trace_create_file("trace", 0644, d_cpu,
3747 (void *) cpu, &tracing_fops);
3749 trace_create_file("trace_pipe_raw", 0444, d_cpu,
3750 (void *) cpu, &tracing_buffers_fops);
3752 trace_create_file("stats", 0444, d_cpu,
3753 (void *) cpu, &tracing_stats_fops);
3756 #ifdef CONFIG_FTRACE_SELFTEST
3757 /* Let selftest have access to static functions in this file */
3758 #include "trace_selftest.c"
3761 struct trace_option_dentry {
3762 struct tracer_opt *opt;
3763 struct tracer_flags *flags;
3764 struct dentry *entry;
3768 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
3771 struct trace_option_dentry *topt = filp->private_data;
3774 if (topt->flags->val & topt->opt->bit)
3779 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3783 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
3786 struct trace_option_dentry *topt = filp->private_data;
3791 if (cnt >= sizeof(buf))
3794 if (copy_from_user(&buf, ubuf, cnt))
3799 ret = strict_strtoul(buf, 10, &val);
3806 /* do nothing if already cleared */
3807 if (!(topt->flags->val & topt->opt->bit))
3810 mutex_lock(&trace_types_lock);
3811 if (current_trace->set_flag)
3812 ret = current_trace->set_flag(topt->flags->val,
3814 mutex_unlock(&trace_types_lock);
3817 topt->flags->val &= ~topt->opt->bit;
3820 /* do nothing if already set */
3821 if (topt->flags->val & topt->opt->bit)
3824 mutex_lock(&trace_types_lock);
3825 if (current_trace->set_flag)
3826 ret = current_trace->set_flag(topt->flags->val,
3828 mutex_unlock(&trace_types_lock);
3831 topt->flags->val |= topt->opt->bit;
3844 static const struct file_operations trace_options_fops = {
3845 .open = tracing_open_generic,
3846 .read = trace_options_read,
3847 .write = trace_options_write,
3851 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
3854 long index = (long)filp->private_data;
3857 if (trace_flags & (1 << index))
3862 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3866 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
3869 long index = (long)filp->private_data;
3874 if (cnt >= sizeof(buf))
3877 if (copy_from_user(&buf, ubuf, cnt))
3882 ret = strict_strtoul(buf, 10, &val);
3888 trace_flags &= ~(1 << index);
3891 trace_flags |= 1 << index;
3903 static const struct file_operations trace_options_core_fops = {
3904 .open = tracing_open_generic,
3905 .read = trace_options_core_read,
3906 .write = trace_options_core_write,
3909 struct dentry *trace_create_file(const char *name,
3911 struct dentry *parent,
3913 const struct file_operations *fops)
3917 ret = debugfs_create_file(name, mode, parent, data, fops);
3919 pr_warning("Could not create debugfs '%s' entry\n", name);
3925 static struct dentry *trace_options_init_dentry(void)
3927 struct dentry *d_tracer;
3928 static struct dentry *t_options;
3933 d_tracer = tracing_init_dentry();
3937 t_options = debugfs_create_dir("options", d_tracer);
3939 pr_warning("Could not create debugfs directory 'options'\n");
3947 create_trace_option_file(struct trace_option_dentry *topt,
3948 struct tracer_flags *flags,
3949 struct tracer_opt *opt)
3951 struct dentry *t_options;
3953 t_options = trace_options_init_dentry();
3957 topt->flags = flags;
3960 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
3961 &trace_options_fops);
3965 static struct trace_option_dentry *
3966 create_trace_option_files(struct tracer *tracer)
3968 struct trace_option_dentry *topts;
3969 struct tracer_flags *flags;
3970 struct tracer_opt *opts;
3976 flags = tracer->flags;
3978 if (!flags || !flags->opts)
3983 for (cnt = 0; opts[cnt].name; cnt++)
3986 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
3990 for (cnt = 0; opts[cnt].name; cnt++)
3991 create_trace_option_file(&topts[cnt], flags,
3998 destroy_trace_option_files(struct trace_option_dentry *topts)
4005 for (cnt = 0; topts[cnt].opt; cnt++) {
4006 if (topts[cnt].entry)
4007 debugfs_remove(topts[cnt].entry);
4013 static struct dentry *
4014 create_trace_option_core_file(const char *option, long index)
4016 struct dentry *t_options;
4018 t_options = trace_options_init_dentry();
4022 return trace_create_file(option, 0644, t_options, (void *)index,
4023 &trace_options_core_fops);
4026 static __init void create_trace_options_dir(void)
4028 struct dentry *t_options;
4031 t_options = trace_options_init_dentry();
4035 for (i = 0; trace_options[i]; i++)
4036 create_trace_option_core_file(trace_options[i], i);
4039 static __init int tracer_init_debugfs(void)
4041 struct dentry *d_tracer;
4044 d_tracer = tracing_init_dentry();
4046 trace_create_file("tracing_enabled", 0644, d_tracer,
4047 &global_trace, &tracing_ctrl_fops);
4049 trace_create_file("trace_options", 0644, d_tracer,
4050 NULL, &tracing_iter_fops);
4052 trace_create_file("tracing_cpumask", 0644, d_tracer,
4053 NULL, &tracing_cpumask_fops);
4055 trace_create_file("trace", 0644, d_tracer,
4056 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4058 trace_create_file("available_tracers", 0444, d_tracer,
4059 &global_trace, &show_traces_fops);
4061 trace_create_file("current_tracer", 0644, d_tracer,
4062 &global_trace, &set_tracer_fops);
4064 trace_create_file("tracing_max_latency", 0644, d_tracer,
4065 &tracing_max_latency, &tracing_max_lat_fops);
4067 trace_create_file("tracing_thresh", 0644, d_tracer,
4068 &tracing_thresh, &tracing_max_lat_fops);
4070 trace_create_file("README", 0444, d_tracer,
4071 NULL, &tracing_readme_fops);
4073 trace_create_file("trace_pipe", 0444, d_tracer,
4074 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4076 trace_create_file("buffer_size_kb", 0644, d_tracer,
4077 &global_trace, &tracing_entries_fops);
4079 trace_create_file("trace_marker", 0220, d_tracer,
4080 NULL, &tracing_mark_fops);
4082 trace_create_file("saved_cmdlines", 0444, d_tracer,
4083 NULL, &tracing_saved_cmdlines_fops);
4085 #ifdef CONFIG_DYNAMIC_FTRACE
4086 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4087 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4089 #ifdef CONFIG_SYSPROF_TRACER
4090 init_tracer_sysprof_debugfs(d_tracer);
4093 create_trace_options_dir();
4095 for_each_tracing_cpu(cpu)
4096 tracing_init_debugfs_percpu(cpu);
4101 static int trace_panic_handler(struct notifier_block *this,
4102 unsigned long event, void *unused)
4104 if (ftrace_dump_on_oops)
4109 static struct notifier_block trace_panic_notifier = {
4110 .notifier_call = trace_panic_handler,
4112 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4115 static int trace_die_handler(struct notifier_block *self,
4121 if (ftrace_dump_on_oops)
4130 static struct notifier_block trace_die_notifier = {
4131 .notifier_call = trace_die_handler,
4136 * printk is set to max of 1024, we really don't need it that big.
4137 * Nothing should be printing 1000 characters anyway.
4139 #define TRACE_MAX_PRINT 1000
4142 * Define here KERN_TRACE so that we have one place to modify
4143 * it if we decide to change what log level the ftrace dump
4146 #define KERN_TRACE KERN_EMERG
4149 trace_printk_seq(struct trace_seq *s)
4151 /* Probably should print a warning here. */
4155 /* should be zero ended, but we are paranoid. */
4156 s->buffer[s->len] = 0;
4158 printk(KERN_TRACE "%s", s->buffer);
4163 static void __ftrace_dump(bool disable_tracing)
4165 static raw_spinlock_t ftrace_dump_lock =
4166 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
4167 /* use static because iter can be a bit big for the stack */
4168 static struct trace_iterator iter;
4169 unsigned int old_userobj;
4170 static int dump_ran;
4171 unsigned long flags;
4175 local_irq_save(flags);
4176 __raw_spin_lock(&ftrace_dump_lock);
4184 if (disable_tracing)
4187 for_each_tracing_cpu(cpu) {
4188 atomic_inc(&global_trace.data[cpu]->disabled);
4191 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4193 /* don't look at user memory in panic mode */
4194 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4196 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4198 /* Simulate the iterator */
4199 iter.tr = &global_trace;
4200 iter.trace = current_trace;
4201 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4204 * We need to stop all tracing on all CPUS to read the
4205 * the next buffer. This is a bit expensive, but is
4206 * not done often. We fill all what we can read,
4207 * and then release the locks again.
4210 while (!trace_empty(&iter)) {
4213 printk(KERN_TRACE "---------------------------------\n");
4217 /* reset all but tr, trace, and overruns */
4218 memset(&iter.seq, 0,
4219 sizeof(struct trace_iterator) -
4220 offsetof(struct trace_iterator, seq));
4221 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4224 if (find_next_entry_inc(&iter) != NULL) {
4225 print_trace_line(&iter);
4226 trace_consume(&iter);
4229 trace_printk_seq(&iter.seq);
4233 printk(KERN_TRACE " (ftrace buffer empty)\n");
4235 printk(KERN_TRACE "---------------------------------\n");
4237 /* Re-enable tracing if requested */
4238 if (!disable_tracing) {
4239 trace_flags |= old_userobj;
4241 for_each_tracing_cpu(cpu) {
4242 atomic_dec(&global_trace.data[cpu]->disabled);
4248 __raw_spin_unlock(&ftrace_dump_lock);
4249 local_irq_restore(flags);
4252 /* By default: disable tracing after the dump */
4253 void ftrace_dump(void)
4255 __ftrace_dump(true);
4258 __init static int tracer_alloc_buffers(void)
4264 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4267 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4268 goto out_free_buffer_mask;
4270 if (!alloc_cpumask_var(&tracing_reader_cpumask, GFP_KERNEL))
4271 goto out_free_tracing_cpumask;
4273 /* To save memory, keep the ring buffer size to its minimum */
4274 if (ring_buffer_expanded)
4275 ring_buf_size = trace_buf_size;
4279 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4280 cpumask_copy(tracing_cpumask, cpu_all_mask);
4281 cpumask_clear(tracing_reader_cpumask);
4283 /* TODO: make the number of buffers hot pluggable with CPUS */
4284 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4285 TRACE_BUFFER_FLAGS);
4286 if (!global_trace.buffer) {
4287 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4289 goto out_free_cpumask;
4291 global_trace.entries = ring_buffer_size(global_trace.buffer);
4294 #ifdef CONFIG_TRACER_MAX_TRACE
4295 max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4296 TRACE_BUFFER_FLAGS);
4297 if (!max_tr.buffer) {
4298 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4300 ring_buffer_free(global_trace.buffer);
4301 goto out_free_cpumask;
4303 max_tr.entries = ring_buffer_size(max_tr.buffer);
4304 WARN_ON(max_tr.entries != global_trace.entries);
4307 /* Allocate the first page for all buffers */
4308 for_each_tracing_cpu(i) {
4309 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4310 max_tr.data[i] = &per_cpu(max_data, i);
4313 trace_init_cmdlines();
4315 register_tracer(&nop_trace);
4316 current_trace = &nop_trace;
4317 #ifdef CONFIG_BOOT_TRACER
4318 register_tracer(&boot_tracer);
4320 /* All seems OK, enable tracing */
4321 tracing_disabled = 0;
4323 atomic_notifier_chain_register(&panic_notifier_list,
4324 &trace_panic_notifier);
4326 register_die_notifier(&trace_die_notifier);
4331 free_cpumask_var(tracing_reader_cpumask);
4332 out_free_tracing_cpumask:
4333 free_cpumask_var(tracing_cpumask);
4334 out_free_buffer_mask:
4335 free_cpumask_var(tracing_buffer_mask);
4340 __init static int clear_boot_tracer(void)
4343 * The default tracer at boot buffer is an init section.
4344 * This function is called in lateinit. If we did not
4345 * find the boot tracer, then clear it out, to prevent
4346 * later registration from accessing the buffer that is
4347 * about to be freed.
4349 if (!default_bootup_tracer)
4352 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4353 default_bootup_tracer);
4354 default_bootup_tracer = NULL;
4359 early_initcall(tracer_alloc_buffers);
4360 fs_initcall(tracer_init_debugfs);
4361 late_initcall(clear_boot_tracer);